Mac OS ARM Native reaches ELF entry #769
Conversation
There was a problem hiding this comment.
Overall this is good enough to get to ARM entry but a bit incomplete and untested, particularly the address space changes could have an impact or might need more modifications. I think you should really make some initial progress with getting an ARM translator running and then PR this stuff once its verified to be a sound base, just to validate that everything here is correct.
CMakeLists.txt
Outdated
| # Reserve system-managed memory space. | ||
| target_link_options(shadps4 PRIVATE -Wl,-no_pie,-no_fixup_chains,-no_huge,-pagezero_size,0x4000,-segaddr,TCB_SPACE,0x4000,-segaddr,GUEST_SYSTEM,0x400000,-image_base,0x20000000000) | ||
| # Using x86_64 (ARM by default if not set) | ||
| if (CMAKE_OSX_ARCHITECTURES) |
There was a problem hiding this comment.
You will need to check more than just CMAKE_OSX_ARCHITECTURES because it may be compiled natively on an Intel Mac. Simplest option I think would be to move the logic added to externals/CMakeList.txt for ffmpeg to this file and use the ARCHITECTURE variable to see if ARCHITECTURE STREQUAL "x86_64"
src/common/assert.cpp
Outdated
| #define Crash() __asm__ __volatile__("int $3") | ||
| #elif __aarch64__ | ||
| #define Crash() ; |
There was a problem hiding this comment.
This should probably have an actual implementation.
src/common/rdtsc.cpp
Outdated
|
|
||
| #else | ||
|
|
||
| return 0; |
There was a problem hiding this comment.
Should have an alternate implementation or at least a TODO note.
| #else | ||
| // ARM | ||
| static inline u64 FencedRDTSC() { | ||
| return 0; |
There was a problem hiding this comment.
Same here for actual implementation or TODO.
| ASSERT_MSG(ret != MAP_FAILED, "mmap failed: {}", strerror(errno)); | ||
| } else { | ||
| int ret = mprotect(reinterpret_cast<void*>(virtual_addr), size, prot); | ||
| ASSERT_MSG(ret == 0, "mprotect failed: {}", strerror(errno)); |
There was a problem hiding this comment.
This and the below in Unmap will need a closer look to make sure the behavior is correct, I put it together for you quick to get things going but I'm not sure it's good enough to go upstream.
There was a problem hiding this comment.
Actually no, this part was @squidbus who has more experience with the memory, so I trust that it's correct. Feel free to test though
[Edit] I just realized it was you that said the message, this was in the mem.patch file you gave me
src/core/linker.cpp
Outdated
| void* Linker::TlsGetAddr(u64 module_index, u64 offset) { | ||
| void* Linker::TlsGetAddr(u64 module_index, u64 offset) | ||
| { | ||
| #ifdef __x86_64__ |
There was a problem hiding this comment.
You should probably just stub GetTcbBase() for now if anything, not the whole thing here.
src/core/linker.cpp
Outdated
| @@ -107,7 +119,9 @@ void Linker::Execute() { | |||
| } | |||
| } | |||
|
|
|||
| #ifdef __x86_64__ | |||
| SetTcbBase(nullptr); | |||
There was a problem hiding this comment.
Instead of stubbing out SetTcbBase calls it should probably just have an stub function to call for now, it's going to need to be set up somehow for ARM translated code to access. Ditto for the other cases of this.
Reaches entry and then stops, the recompiler comes next